I’m Kristef. I turn messy data and research questions into explainable software that people can actually use.
My sweet spot is where rigorous research meets working products. I care about what a system finds, how it explains itself, and whether the people using it can trust the result.
(01—05)01 / SELECTED WORK
Projects spanning industry research, data engineering, machine learning, and human-centered design.
Software vacancies say more than which tools a role requires. They also communicate responsibility, culture, working conditions, and expectations. I built a way to examine those signals across vacancies while keeping every finding connected to its source text.
Explore the case
How can a company compare what its software vacancies communicate?
Built a literature-informed signal framework and applied its frozen rules to 12 vacancies.
Created a React/FastAPI prototype that links detected signals to the original wording.
Studied wording with 27 participants and gathered formative feedback from 9 CPP stakeholders.
A SMALL INTERACTIVE EXAMPLE
Select a signal to see the evidence highlighted in a sample vacancy sentence.
Illustrative text created for this portfolio. The thesis prototype supports human review; it does not score candidates or make hiring decisions.
At TU/e, I helped turn scattered Devpost records into a linked dataset and interactive Streamlit dashboard. The pipeline handled dynamic pages, cleaned and connected entities, and made large-scale patterns explorable.
LEARNING CURVE / LASER BEAM IMAGESFor an industrial beam-shape classification task, our team explored active learning to use a limited labeling budget. We also tested probability calibration, comparing Platt scaling and isotonic regression.
We implemented sampling strategies and tracked test accuracy as the labeled set grew. The plot shows the learning trajectory from a small initial set.
We examined whether predicted probabilities matched observed outcomes, using expected calibration error and visual calibration curves.
Coursework evaluation on small datasets; these results are not a production validation of the model.
MORE WAYS I WORK
A CNN–LSTM pipeline for generating descriptions from visual input. B.Tech capstone work presented at IRCCTSD ’24 and published in conference proceedings.
Python analysis and Tableau views to make patient-reported trigger patterns easier to inspect, including temporal and correlation views.
02 / HOW I THINK
Good technical work needs a clear path from the original question to the result someone sees on screen.
Find the decision or task behind the data. Define what the system can say and where its evidence ends.
Keep assumptions, methods, and source evidence visible. Let people inspect and challenge the output.
Prototype the interaction, gather feedback, and improve the parts that slow understanding down.
03 / A LITTLE ABOUT ME
CURIOUS BY NATURE.I’m a computer science graduate from Eindhoven University of Technology, with a background in software engineering, empirical research, data analysis, and interactive systems. I like work that asks both “does it function?” and “does it help someone understand?”
My path has taken me from accessibility-focused deep learning and health data visualization to a large-scale data product and an industry thesis at Canon Production Printing. Across them, I’ve learned to move between code, evidence, design, and feedback.
04 / LET'S CONNECT
I’m interested in software, data, and applied AI roles where research can become a useful product.